Estimation of Permeability and

Permeability Anisotropy From

Straddle-Packer Formation-Tester
Measurements Based on the Physics of
Two-Phase Immiscible Flow and Invasion
Renzo Angeles, Carlos Torres-Verdn, Hee-Jae Lee, and Faruk O. Alpak, University of Texas at Austin; and
James Sheng,* Baker Atlas

Summary
We describe the successful application of a new method to estimate permeability and permeability anisotropy from transient measurements of pressure acquired with a wireline straddle-packer
formation tester. Unlike standard algorithms used for the interpretation of formation-tester measurements, the method developed in
this paper incorporates the physics of two-phase immiscible flow
as well as the processes of mudcake buildup and invasion.
An efficient 2D (cylindrical coordinates) implicit-pressure explicit-saturation finite-difference algorithm is used to simulate
both the process of invasion and the pressure measurements acquired with the straddle-packer formation tester. Initial conditions
for the simulation of formation-tester measurements are determined by the spatial distributions of pressure and fluid saturation resulting from mud-filtrate invasion. Inversion is performed
with a Levenberg-Marquardt nonlinear minimization algorithm.
Sensitivity analyses are conducted to assess nonuniqueness and
the impact of explicit assumptions made about fluid viscosity,
capillary pressure, relative permeability, mudcake growth, and
time of invasion on the estimated values of permeability and permeability anisotropy.
Applications of the inversion method to noisy synthetic measurements include homogeneous, anisotropic, single- and multilayer formations for cases of low- and high-permeability rocks. We
also study the effect of unaccounted impermeable bed boundaries
on inverted formation properties. For cases where a priori information can be sufficiently constrained, our inversion methodology
provides reliable and accurate estimates of permeability and permeability anisotropy. In addition, we show that estimation errors
of permeability inversion procedures that neglect the physics of
two-phase immiscible fluid flow and mud-filtrate invasion can be
as high as 100%.
Introduction
Modular and multiprobe formation testers have proved advantageous in the determination of permeability at intermediate-scale
lengths because of the increased distance between the observation
and sink probes (Pop et al. 1993; Badaam et al. 1998; Proett et al.
2000). Moreover, the use of dual-packer or straddle-packer modules over point-probe modules is known to improve the interpretation of pressure transient measurements when testing laminated,
shaly, fractured, vuggy, unconsolidated, and low-permeability formations (Ayan et al. 2001). Several papers have been published to
describe interpretation techniques and applications of these new
formation-testing approaches (Kuchuk 1998; Hurst et al. 2000;
Onur et al. 2004).

* Currently with TOTAL.

Copyright 2007 Society of Petroleum Engineers
This paper (SPE 95897) was first presented at the 2005 SPE Annual Technical Conference
and Exhibition, Dallas, 912 October, and revised for publication. Original manuscript received for review 14 July 2005. Revised manuscript received 22 May 2007. Paper peer
approved 29 May 2007.

September 2007 SPE Journal

The new method introduced in this paper interprets formationtester measurements acquired with wireline straddle-packer tools.
It incorporates the physics of two-phase, axisymmetric, immiscible
fluid flow to simulate the measurements, and it is combined with
a nonlinear minimization algorithm for history-matching purposes.
Comparable inversion approaches have been documented in the
open technical literature (Proett et al. 2000; Xian et al. 2004;
Jackson et al. 2003) but they assumed single-phase fluid flow.
Recently, Zeybek et al. (2001) introduced a multiphase flow
method to integrate formation-tester pressure and fractional flow
measurements with the objective of refining relative permeability
values estimated from openhole resistivity logs. The same authors
considered the manual inversion of radial invasion profiles, horizontal permeability, and permeability anisotropy but did not assess
the uncertainty of their estimations introduced by a priori assumptions about multiphase flow parameters. By contrast, the developments reported in this paper integrate the flow simulator with a
dynamically coupled mudcake growth and mud-filtrate invasion
algorithm (Wu et al. 2002), which improves the physical consistency and reliability of the quantitative estimation of both permeability and permeability anisotropy.
Method
There are three main components in the workflow developed in
this paper:
1. Mud-filtrate invasion algorithm
2. Two-phase axisymmetric simulator
3. Nonlinear minimization algorithm
Transient measurements of pressure and flow rate are compared to the outputs of a two-phase axisymmetric simulator to
yield new model parameters through nonlinear minimization. The
invasion algorithm makes use of these parameters (permeability
and permeability anisotropy), in addition to pressure overbalance,
invasion geometry, mudcake properties, and other rock-formation
properties, to simulate the process of mud-filtrate invasion. Subsequently, the calculated spatial distributions of pressure and fluid
saturation resulting from mud-filtrate invasion are used as initial
conditions for pressure-transient tests. To reduce the time required
by the inversion, this last step could be approximated with an
invariant mud-filtrate invasion profile calculated only once during
the minimization. However, such a strategy is not recommended
for supercharged formations where updates of initial conditions
during minimization can drastically impact inversion results. For
the synthetic case examples considered in this paper, we constantly
update the initial conditions during minimization. The geometry of
the formation model (e.g., multilayer formations, impermeable bed
shoulders) is fixed when entered into the flow simulator. Consequently, field measurements, well-log measurements, and other
independent sources of information are needed to define the geometrical properties of the rock formation model. To complete the
estimation, the fluid-flow simulator yields pressure transients to be
compared against actual measurements. This process repeats itself
until the quadratic norm of the residuals between simulations and
measurements decreases to a predefined value. When the latter
339

Fig. 1Configuration of the base-case straddle-packer, a

wireline formation tester consisting of two vertical observation
probes and a dual-packer module.

Our base-case model replicates the conditions of an invaded

zone through the injection of brine into the formation during 1.5
days. This value, as well as other assumptions on rock formation
and mud properties, was arbitrarily chosen to illustrate the method
proposed in this paper rather than to describe a specific situation.
Additional assumptions include the values of brine salinity, equal
to 5,000 ppm (1.75 lbm/STB) and formation water salinity, equal
to 120,000 ppm (42.06 lbm/STB). Table 1 summarizes the properties of the assumed mud (and mudcake). Fig. 2 describes the assumed water/oil relative permeability and capillary pressure curves.

condition is met, the inversion algorithm outputs the estimated

values of permeability and permeability anisotropy.
Systematic description of the inversion methodology on synthetic measurements requires the use of a base case model (described later), which includes petrophysical variables and geometrical properties that can reproduce arbitrary formation models
and tool configurations. Fig. 1 shows the measurement configuration for the assumed straddle-packer wireline formation tester.
Dimensions of the base case tool were chosen according to
typical physical dimensions of wireline formation testers designed
for interval pressure transient tests (IPTTs): two vertical observation probes and one packer flow area that acts as a sink. The
simulator used in this paper was developed by the Formation
Evaluation Research Program of the University of Texas at Austin.
Special considerations about skin factor, tool storage, and total
compressibility are discussed in a separate section of the paper.
Although not investigated here, skin damage can be readily implemented in the inversion method by using a similar approach to the
multilayer formation example presented in this work. In addition,
even though we ignore tool-storage effects, the latter can be studied with time-variable flow rates of fluid production.
Numerical Simulation of the Process of
Mud-Filtrate Invasion
An adaptation of INVADE, developed by Wu et al. (2002), is used
to simulate the process of mud-filtrate invasion. Simulations include the dynamically coupled effects of mudcake growth and
multiphase, immiscible mud-filtrate invasion. In simple terms, the
flow rate of mud-filtrate invasion depends on both mud and rock
formation properties. This approach differs from the procedure
described by Gk et al. (2006), who considered stationary composite zones and assumed single-phase flow. By coupling the invasion algorithm with the flow simulator, our inversion method is
not restricted to discontinuous fluid saturation gradients and, more
importantly, it does not assume that the invaded zone across the
straddle-packer interval is immobile nor stays under constant fluid
saturation during the test (i.e., our inversion approach remains
accurate in cases of significant fluid cleanup). The INVADE algorithm assumes that the rock formation is drilled with a waterbased mud (WBM).
340

Fig. 2Water/oil relative permeability and capillary pressure

curves assumed in the numerical simulations of both mudfiltrate invasion and formation-tester measurements.
September 2007 SPE Journal

Fig. 3Graphical description of the finite-difference grid used

for the numerical simulations associated with the base-case
model. The reservoir extends from 0.354 ft (wellbore radius) to
1,000 ft (drainage radius) horizontally and from 3,990 ft to 4,020
ft vertically. The numbers located to the left of the tool schematic indicate the distance in feet from each probe and packer
to the top of the reservoir. The total pay thickness is 30 ft.

Fig. 4Graphical description of the finite-difference grid used

for the study of impermeable bed boundaries. Numbers located
to the left of the tool schematic indicate the distance in feet from
each probe and packer to the top of the reservoir. The numerical
simulations assume negligible values of permeability and porosity outside the permeable layer.

To couple the outputs of the invasion algorithm with the numerical simulation of pressure transient measurements, the simulator calculates spatial distributions of pressure, salt concentration,
and fluid saturation resulting from 1.5 days of invasion that are
entered as initial conditions for the simulation of formation-tester
measurements.

difference grid used in this paper to assess the effects of impermeable bed boundaries on formation-tester measurements.
There are three observation points for the measurement of pressure transients at distances of 5, 13, and 20 ft measured from the
top of the reservoir, respectively. The packer interval (sink) has a
length of 2 ft. Upper, lower, and external reservoir boundaries are
assumed impermeable (flow rate is zero). Table 2 summarizes the
geometrical dimensions of the reservoir model considered in this
section, whereas Table 3 describes the associated rock and fluid
properties. Initial conditions for formation-tester measurements
prior to the onset of mud-filtrate invasion are given in Table 4. For
the base-case model, the drawdown sequence enforces a constant
production flow rate at the packer of 21 STB/D during 60 minutes,
after which the buildup sequence continues for 60 additional minutes. Fig. 5 illustrates the assumed flow-rate sequence.
To validate the finite-difference grid used in this paper, we
conducted the test shown in Fig. 6, where we compared packerpressure measurements simulated for the base case model against
the corresponding single-phase radial-flow analytical Ei solution.
The rock system is anisotropic in order to emphasize radial flow
conditions in the comparison exercise. Results indicate an excellent match between the numerical and analytical results.

Numerical Simulation of Two-Phase Flow

The simulation of formation-tester measurements is performed
with a water/oil two-phase, axisymmetric reservoir simulator developed by the Formation Evaluation Program of the University of
Texas at Austin. Detailed information about this simulator is given
by Lee et al. (2004). The simulator uses finite differences and the
IMPES (implicit-pressure explicit-saturation) algorithm to solve
the nonlinear system of equations of pressure and saturation. Various boundary conditions can be enforced by the algorithm. Spaceand time-dependent variables such as temperature and salt concentration are also included in the simulations.
Base-Case Model. The Base-Case Model describes standard measurement parameters and rock formation properties assumed for
most of the studies considered in this paper. Fig. 3 shows the
finite-difference grid along with the physical dimensions of the
assumed hydrocarbon-bearing rock formation. The vertical separation between grid nodes is nonuniform, ranging from 0.5 ft near
the probes to 1 ft at the remaining grid nodes. In the radial direction, the simulator enforces a logarithmic discretization, starting
from an initial value of 0.049 ft near the wellbore to 122.2 ft at the
outer radial boundary of the reservoir. Fig. 4 describes the finite-

September 2007 SPE Journal

Nonlinear Inversion Algorithm

Given the complex nonlinear relationship between borehole pressure-gauge measurements and rock formation petrophysical prop-

341

erties, the inversion algorithm requires several sequential linear

steps to match the simulated pressure transients with measured
pressure transients. Similar types of applications can be found in
the open technical literature (Kuchuk 1998; Onur et al. 2004).
The inversion algorithm considered in this paper is based on the
general framework for constrained minimization described by Habashy and Abubakar (2004). Specifically, we use a modified version of the Levenberg-Marquardt (LM) minimization method
implemented by Alpak (2005). The LM minimization method is
widely used for nonlinear least-squares problems because of its
stability and good rate of convergence. Away from the minimum,
the algorithm is similar to the steepest-descent method, whereas in
the neighborhood of the minimum, the algorithm is similar to the
Gauss-Newton method. Accordingly, model parameters are obtained by minimizing a cost function composed of the quadratic
misfit between measured and simulated samples of transient pressure plus an additive quadratic model functional. For this work, the
cost function is written as
1
Cx = ex 2 + x 2, . . . . . . . . . . . . . . . . . . . . . . . . . . . (1)
2
where , the Lagrange multiplier, is a scalar quantity (0<<) that
assigns relative weight to the two additive terms included in Eq. 1,
and e(x), the vector of data residuals, is defined below. The first
additive term of the cost function yields an estimate of the unknown model x that honors the measurements, whereas the second
additive term prevents instability in the estimation due to nonuniqueness as well as insufficient and noisy measurements. Selection of the Lagrange multiplier is based on the criteria
1
= maxmm,

provided that

minmm
,
maxmm

where is a constant equal to 10 and m are the eigenvalues of

the real and symmetric matrix JT(x)J(x). The Jacobian matrix,
J(x), is given by

ments is used to identify the corresponding time sample. In the

above expression, data residuals are normalized in order to put
both packer and probe pressure transients on equal footing. An
alternative approach is to define the measurements as pressure
differentials (i.e., pjppj) where p is formation pressure prior
to the time of measurement, or to use the logarithmic value of the
pressure differentials, log(pj), in which case the normalization of
Eq. 2 is no longer required. In this paper, total pressures were used
to obtain the inversion results reported in Tables 5 and 6 except
for those values identified with (*), where pressure differentials
were necessary to ensure stability of the inversion. Pressure differentials were also used to obtain the results reported in Table 7.
Our experience shows that, in general, the use of pressure differentials substantially increases the stability of the minimization process. Further comparisons between the use of relative misfit errors
and pressure differentials to perform the inversion can be found in
the work by Angeles (2005).
The vector of model parameters included in Eq. 2, x, is given by
x=

Jx= Jmn =

Fig. 5Time schedule of flow rate assumed for the simulations

of formation tester measurements in connection with the base
case model.

logy1

, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (3)

logyN

x1

xN

em
, m = 1, 2, 3, . . . , M; n = 1, 2, 3, . . . , N ,
xn

where M is the number of measurements, and N is the number of

unknown model parameters. The weight of the misfit term in the
cost function is progressively adjusted by the Lagrange multiplier
as the inversion algorithm iterates toward the minimum of the cost
function. This approach guarantees a stable intermediate solution
at every iteration without over-biasing the final solution by the
specific choice of regularization term.
The inversion algorithm is based on the following definition of
vector of data residuals e(x):

ex =

e1x

ej x

eM x

ps1x pm1 pm1

psj x pm j pm j

psMx pmM pmM

, . . . . . . . (2)

where pmj is the measured pressure and psj is the numerically

simulated pressure. The index j attached to the pressure measure342

Fig. 6Validation test of the finite-difference grid employed to

simulate the base-case model. Three different formation permeability values (10, 100, and 1000 mD) are the basis of comparison between simulated packer-pressure measurements and the
single-phase radial-flow analytical solution.
September 2007 SPE Journal

where N is the number of unknowns, and yi is permeability. For the

purposes of this paper, model parameters are either permeability or
permeability anisotropy ratio. When inverting for permeabilities,
logarithmic values are used to define the model parameters yi, as
September 2007 SPE Journal

indicated in Eq. 3. On the other hand, when inverting for permeability anisotropy, actual (nontransformed) values are used instead.
The logarithmic transformation of permeability enhances the convergence rate of the inversion algorithm (Angeles 2005). When the
343

inversion is implemented to estimate mobilities, the corresponding

viscosities of the fluids involved are assumed constant.
Although not implemented in this paper, the quadratic cost
function defined by Eq. 1 could include a data weighting matrix,
Wd, to de-emphasize the influence of noisy or biased pressure
samples. In such a case, the term e(x)2 becomes Wd e(x)2, and
if the measurement noise is stationary and uncorrelated, then
Wddiag{1/j}, where j is the standard deviation of the noise
present in the jth pressure measurement.
Cramr-Rao Uncertainty Bounds. We use the Cramr-Rao
(Cramer 1945) approach to estimate the uncertainty of the estimated parameters (permeability and permeability anisotropy). Accordingly, a perturbation is performed on the parameters yielded
by the inversion to evaluate the expression
22I + JTx*Jx*1, . . . . . . . . . . . . . . . . . . . . . . . . . . (4)
where is the standard deviation of the uncorrelated, zero-mean
Gaussian noise used to contaminate the pressure data, is the
Lagrange multiplier included in the quadratic cost function (Eq. 1),
x* is the vector of inverted model parameters, and is the estimators covariance matrix. The square root of the diagonal entries
(variances) of the latter matrix provide the uncertainty values such
that for 99.7% of the occurrences, the ith model parameter will fall
within 3ii of the estimated value (Habashy and Abubakar
2004). In this paper, uncertainty bounds were calculated only for
the case of noisy measurements for inversion examples of multilayer rock formations; nonetheless, uncertainty bounds could also
be calculated for all noisy synthetic cases considered the paper.
344

Data Misfit and Impact Value. In addition to the estimated parameters, there are two diagnostic outputs provided by the inversion process: data misfit and impact value. Data misfit is quantified
with the root mean square (RMS) difference between the input and
simulated transient pressure measurements at the end of the inversion. This value is given as a percentage of the quadratic norm of
the input transient pressure measurements. For noise-free synthetic
cases, the data misfit is expected to be 0.0%.
We introduce an impact value (IM) to quantify the relative
importance of specific assumptions made in the inversion process
(e.g., fluid viscosity, irreducible water saturation, and level of
noise). The IM is defined as
IM = 100

l
, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (5)
max

with

l =

1
L1

l=1

12

x*
l xl

. . . . . . . . . . . . . . . . . . . . . . . . . . (6)

In these expressions, the subscript l identifies the specific inversion

result for a given inversion example, xl* is the estimated value
obtained from the inversion, x
l is its corresponding target value (or
the actual value for synthetic examples), and max is the maximum
value of l in the total inversion scheme for single-layer inversion
examples (as reported in Tables 5 and 6) or for multilayer inversion
examples (as reported in Table 7). For illustration, consider the case of 10
psi noise in the drawdown only section of Table 5. Accordingly,
September 2007 SPE Journal

10.1 10 + 101.9 100 1 2

= 142.06.
+ 1200.9 1000
For single-layer inversion examples, max was found to be 2,837.3;
hence, the corresponding impact value is IM100*(142.06)/
2,837.35.01. Impact values range from 0 to 100, the closer to
100 the largest the influence of a given assumption on the inverted
parameters. Conversely, impact values close to 0 indicate the lowest possible influence on inverted parameters. The impact value
is provided here for qualitative interpretation of the sensitivity
studies and is influenced by the specific inversion examples
considered within a given set of inversion results. Similar diagnostic procedures could be implemented in field applications by
changing the target value x
l to the final matched value and by
calculating the variations of x l* for specific assumptions on formation properties.

one has l = 1 2*

Considerations on Damage Skin, Tool Storage,

and Compressibility
The synthetic cases considered in this paper do not explicitly include the effects of skin, tool storage, or total fluid compressibility
on pressure-transient measurements. However, it becomes important to illustrate how these parameters can be readily incorporated
into the inversion method described in this paper.
To include skin factor in the inversion, we suggest constructing
a composite radial model where a damage or stimulated layer
adjacent to the wellbore is assigned a radius (rs) and a permeability
(ks), both of which could be regarded as unknown parameters in
the inversion. Subsequently, skin (s) could be estimated with
Hawkins formula (1956):
s=

rs
k
1 ln , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (7)
ks
rw

where rw is the wellbore radius. Alternative approaches are possible such as those described by Pucknell and Clifford (1991).
However, we anticipate that the increased number of unknowns
will worsen the stability of the inversion, therefore requiring additional information to guarantee reliable estimations.
To include tool storage, the simulator (as with most reservoir
simulators) can take as input time-variable flow rates of fluid
production. In cases of severe compression and decompression of
fluid across the flowline, data processing techniques (e.g., lowpass filtering) applied to the measured flow rates could help to
improve the estimation. Note, however, that these propositions
have not been explored by the authors.
September 2007 SPE Journal

To include total compressibility, one could add as unknown

parameters both water (cw) and oil (co) compressibilities. Given
that fluid saturations are output by the flow simulator, it is possible
to calculate averaged values of water and oil saturations (S w and
S o, respectively). Accordingly, the average total compressibility c t
is given by
c = S c + S c + c , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (8)
t

w w

o o

where cf is the formation rock compressibility.

Sensitivity of Straddle-Packer Formation Tester
Measurements to Rock-Formation Properties
This section evaluates the impact of specific assumptions made
about rock-formation properties on the simulated pressuretransient measurements and the corresponding inversion results.
For this purpose, the base-case model (described earlier) is used to
establish the standard measurement parameters and rock-formation
properties assumed for all the examples under consideration. Explicit correlations between permeability, porosity, capillary pressure, and relative permeability are enforced by the inversion algorithm as described in the Appendix. The flow rate of mud filtrate
was restricted to a maximum of 0.003 B/D/ft for the examples
presented in this section, except for the sensitivity analyses of
mud-filtrate invasion, where flow rates exceeded 0.2 B/D/ft.
Refer to Tables 5 through 7 for a summary of the results obtained in this section. With the exception of the cases identified as
Only Drawdown and Impermeable Bed Boundaries, all the
results presented in these tables were obtained using the simultaneous inversion of the two monitoring probes and the packer as
well as both drawdown and buildup pressure sequences.
Sensitivity to Variations of Permeability. Four different types of
rock formations are used to assess the effect of the spatial distribution
of permeability and permeability anisotropy on inversion results:
1. Homogeneous and isotropic formations
2. Homogeneous and anisotropic formations
3. Two-layer and isotropic formations
4. Finely laminated formations
Homogeneous and Isotropic Formation. Fig. 7 describes the
pressure responses simulated at two of the three sampling points of
the formation tester: monitoring probe 1 and packer. It is observed
that the simulated pressure-transient measurements for highpermeability rocks exhibit relatively lower sensitivity to variations
of rock permeability than those associated with low-permeability
rocks (i.e., small variations of permeability cause small variations
345

Fig. 8Log-log plot for the packer and probe 2 buildup pressure measurements simulated for the base-case model. Simulated measurements for the monitoring probe 1 are not shown
here but are also used for inversion. In addition, the plot compares single- and two-phase flow synthetic measurements. The
permeability of the formation is 100 mD.

these pressure samples is the same as the one enforced by the

numerical simulator. Each iteration requires approximately 3.5
minutes of CPU time on a 1.6-GHz Windows-based PC. The inversion reached convergence within nine iterations. Final inverted
permeabilities are the same as the target values.

Fig. 7Noise-free simulated pressure transient measurements

at the observation probe and packer located at 5 and 20 ft,
respectively, from the top of the reservoir.

of pressure only when permeabilities are high). For instance, a

variation from 100 to 200 mD modifies the late-time packer pressure response by 5%. On the other hand, an increase from 900 to
1,000 mD only modifies it by 0.01%. This effect is further enhanced when the measurements are artificially contaminated with
zero-mean additive Gaussian noise.
Another interesting observation from Fig. 7 is the range of
variation of pressure measured at both packer and probes. Probepressure variations are given in a few fractions of psi, whereas
packer responses vary in the range of psi or tens of psi. For example, a pressure drawdown of 7.7 psi is simulated for a 10-mD
rock formation at the probe. This represents 6.7% of the drawdown
pressure (114.8 psi) simulated at the packer for the same formation.
Fig. 8 describes the flow regimes observed in the base-case
model. At early times, the log-log plot exhibits the 1/2 slope
characteristic of spherical flow (before approximately 0.001 hours)
but subsequently stabilizes to a constant value (i.e., a radial flow
regime is reached after approximately 0.01 hours of buildup for
both monitoring and packer-pressure measurements). Table 5
shows the inverted values of permeability starting with an initial
guess of 40 mD toward their respective target values. In this case,
the inversion algorithm is driven by noise-free synthetic pressure
transient data generated with the two-phase flow simulator at both
probes and the packer. A total of 2,592 time pressure samples from
the complete test interval (including drawdown and buildup) were
used by the inversion algorithm. The time sampling used to acquire
346

Effect of Additive Zero-Mean Random Gaussian Noise. For

this exercise, packer and probe measurements were equally contaminated with zero-mean additive random Gaussian noise of standard deviations equal to 1 and 10 psi. Pressure responses associated with high-permeability formations are the most affected by
the presence of noise, especially at the observation probes, where
pressure variations are smaller in amplitude than those at the
packer flow area. As permeability increases, the error increases to
20% for the case of 10-psi Gaussian noise (Table 5). An explanation for this is that high-permeability formations entail smaller
pressure differentials and hence are more susceptible to the presence of noise than low-permeability formations. For instance, a
variation of 1 psi of the measured transient pressure for a formation of 1,000-mD represents a contamination of 60% of the measurements, and this causes an error equal to 1% in the estimated
permeability. Correspondingly, a variation of 1 psi represents a
contamination of 8% for 100-mD pressure measurements, and this
causes a zero error on the estimated permeability. Fig. 9 describes
the misfit between the estimated and measured pressure values.
Effect of Buildup and Drawdown Measurements. In order to
assess the relative information content of each stage of the pressure
transient test, two inversion schemes were designed: one using
only the drawdown pressure measurements, and another using both
drawdown and buildup pressure measurements. Both cases assumed noise-free and noisy synthetic data (10 psi standard deviation of zero-mean additive Gaussian noise). As expected, both test
stages (buildup+drawdown) contribute to decrease the nonuniqueness of the inversion with respect to the case of only one stage
(drawdown in this case), especially when the data are contaminated with relatively large amounts of noise (i.e., 10 psi standard
deviation zero-mean additive Gaussian noise). Although not
shown here, the convergence rate significantly improves when
using both measurement stages (Angeles 2005).
Homogeneous and Anisotropic Formation. Three values of
anisotropy ratio (defined as the ratio between horizontal and vertical permeability, kh/kv) were considered for the base-case rock
formation model: 1, 10, and 100. Fig. 10 shows the corresponding
September 2007 SPE Journal

Fig. 9Comparison of input pressure measurements and pressure measurements simulated with the permeability estimated
from inversion. Input pressure measurements were simulated
for a formation with homogeneous and isotropic permeability
equal to 100 mD, and they were contaminated with additive
zero-mean random Gaussian noise of standard deviation equal
to 10 psi. The inverted permeability is equal to 101.6 mD.

simulated pressure measurements. From the figures, one can observe that an increase of anisotropy ratio causes an increase of the
magnitude of the pressure differential at the packer flow area.
Conversely, the pressure differential decreases at the vertical observation probes. This effect is emphasized for the case of highpermeability formations, where the corresponding pressure measurements are much smaller in magnitude than those associated
with low-permeability formations. It then follows that the pressure
measurements acquired at the probes might not contribute significantly to reduce nonuniqueness of the inversion if the formation
permeability is high (above 500 mD for the examples shown in this
paper). In the latter case, alternative pressure-testing strategies
could be used to reduce nonuniqueness of the inversion (e.g., by
acquiring pressure measurements at two different depths). This is
an important technical issue because pressure measurements are
often corrupted by noise, thereby decreasing their reliability to
estimate rock formation properties.
Unlike the strategy adopted for one-parameter inversion, the
algorithm was allowed to vary the unknown model parameters
within narrower prescribed bounds, thereby reducing the problem
of nonuniqueness discussed previously. For instance, if the inversion algorithm was used to estimate a horizontal permeability
of 100 mD, the minimization would be constrained to find a solution between the upper and lower bounds of 1 and 300 mD,
respectively, compared to the upper and lower bounds of 0.1 and
10,000 mD, respectively, usually enforced for the case of oneparameter inversions.
The inversion algorithm required between 5 to 25 iterations to
achieve convergence toward the estimated parameters. Compared
to the typical range of 5 to 10 iterations normally used for oneparameter inversions, the required number of iterations is relatively high. The algorithm used 552 time-pressure samples acquired with the same time sampling interval used by the simulator.
Also, it was observed that several combinations of permeability
and permeability anisotropy could lead to local minima. This observation indicates that there are several equivalent solutions to the
inverse problem that honor the measurements.
Different inversion techniques were implemented to obtain the
results described previously. Using a priori knowledge, the initial
guess parameters were given values close to their actual values.
In practical applications, such information could be derived
from well-log, core, or production data. When this a priori information was not adequate, the algorithm would continuously restart
September 2007 SPE Journal

the search with different initial guesses to warrant stable convergence while enforcing the same bounds to explore several local
minima. The final result was chosen as the one that entailed the
lowest data misfit.
However, two problems remain for the case of low-permeability formations: first, more local minima exist than for the case
of high-permeability formations (for the same range of anisotropy
ratios, a much larger combination of formation permeabilities honors similar pressure measurements), thereby biasing the inversion
toward values close to the initial guess, and second, estimated
values would converge toward the correct values, but the rate of
convergence was slow. To circumvent these two problems, the
inversion algorithm was modified to use pressure differentials
(pppmj) rather than raw pressure measurements. This strategy proved efficient to reduce nonuniqueness in the inversion results. The corresponding inversion result is identified with an asterisk (*) in Table 5.
The impact of noise contamination on the input pressuretransient measurements is also shown in Table 5. Because of the
severe nonuniqueness of the inverse problem, the algorithm used
several restart values to yield the final estimates. Fig. 11 is a
log-log plot that compares the inverted pressure measurements
against the original noisy measurements generated with zero-mean
additive random Gaussian noise of standard deviation equal to 1
psi. For this example, the inversion algorithm yielded values of
permeability and permeability anisotropy equal to 100.8 mD and
110, respectively, compared to target values of 100 mD and 100,
respectively. Results are deemed satisfactory.
Multilayer Formations
Two-Layer Formation. Fig. 12 describes the examples of twolayer formations considered in this section. We position two of the
observation probes within the top layer and locate the packer
within the bottom layer. Both layers are assumed homogeneous
and isotropic. Also, we assume that layer boundaries as well as
distances between the packer and probes are known a priori. Measurements were contaminated with zero-mean additive random
Gaussian noise of standard deviation equal to 0.1 psi. Simulated
pressure differentials (p) were entered to the inversion algorithm
instead of raw pressure measurements to mitigate the problem of
nonuniqueness. We also calculate Cramer-Rao uncertainty bounds
(using the +/ operator) to quantify the level of confidence of the
estimated properties. Table 7 describes the results of this inversion
exercise. Even though both Cases A and B were initialized with a
guess of 300 mD, we observe that Case A led to convergence in the
first attempt, as opposed to Case B, wherein the inversion stagnated at a local minimum. Such a behavior prompted us to restart
the inversion several times while pursuing the global minimum.
The same strategy was successfully applied to cases of noisecontaminated measurements.
Another important observation from this inversion exercise is
that the Cramer-Rao bounds decrease when inverting properties of
bottom layers. This behavior indicates that the best estimates correspond to zones closer to the packer, where pressure transients are
more sensitive to rock-formation properties.
Finely Laminated Formation. A different inversion methodology
is adopted for the case of finely laminated rock formations. Fig. 13
shows the example of a formation model composed of seven homogeneous and isotropic layers. Testing of this formation model is
performed within the lowest pay zone, where the packers are located, while the vertical observation probes sample pressures
within the medium and top pay zones. Notice that the vertical
separation of the numerical grid nodes is 0.5 ft near the sampling
points, while the thickness of the formation layers varies from 1.5
ft (top) to 2 ft (medium and bottom). Only the three pay-zone
permeabilities are assumed unknown in the estimation. Flow-rate
schedules and formation properties are the same as those assumed
for the base-case formation model.
In this example, the inverse problem is severely nonunique, and
therefore a priori information is necessary to estimate the location
of layer boundaries and the initial guess permeabilities. Moreover,
347

Fig. 10Simulated pressure transient measurements for three values of permeability anisotropy (=kh /kv): 1, 10, and 100 of a rock
formation with horizontal permeability (kh) equal to 100 mD.

the restart strategy proves to be insufficient if it is not combined

with the enforcement of physical bound constraints on the estimated parameters, also assumed to be known a priori. The results
shown in Table 7 were obtained with an initial guess of 120, 250,
and 150 mD for the permeabilities of top, center, and bottom pay
zones, respectively. Table 8 describes the bound constraints enforced for this example. Note that in field applications, one would
preferentially position the packer within more than one pay zone
and acquire additional sets of pressure data to increase the confidence on these bound constraints. We observe that the Cramer-Rao
bounds indicate more confidence on the estimated bottom payzone permeability (located across the packer) than on the permeabilities of the remaining pay zones.
Sensitivity to Variations of Fluid Viscosity
Given the two-phase nature of the fluid-flow phenomenon assumed in this paper, sensitivity analyses were performed to evalu348

ate the impact of assumptions made on the viscosity of the fluids

involved (oil and water for the base-case formation model). Pressure transient measurements were simulated for values of oil viscosity equal to +500% and 20% of the original viscosity. The
same relative variations are applied for the appraisal of water
viscosity. It was observed that the impact of oil viscosity is much
more significant than that of water viscosity. Pressures change
dramatically when oil viscosity is perturbed from its original value.
On the other hand, water viscosity affects only the initial pressure
measurements during drawdown. This behavior can be explained
by recalling the plots of water saturation, relative permeability, and
capillary pressure. Specifically, as the drawdown stage begins,
water saturation decreases from almost 0.63 to 0.45 (i.e., within
a region where the relative permeability of water is not as high
as that of oil). Thus, the effect of water is noticeable only at the
start of drawdown. An increase of oil viscosity decreases the
mobility, and the corresponding effect on pressure is similar to
September 2007 SPE Journal

Fig. 11Log-log plot comparing the buildup measured noisy

synthetic data against simulation results for one of the inversion cases with anisotropy (dotted lines). Although used in the
inversion, pressure measurements acquired with the monitoring probe 1 are not shown here. The agreement between the two
sets of measurements is excellent despite the high level of noise.

that of single-phase flow. Flow-rate schedule and remaining formation properties were the same as those of the base-case formation model.
Sensitivity to Variations of Capillary Pressure
and Relative Permeability
Table 5 also shows the effects of variations of the assumed water/
oil capillary pressure and relative permeability curves on inversion
results. For this purpose, the study makes use of modified BrooksCorey parameters as well as base-case model properties. Sensitivity analyses consider variations of water/oil capillary pressure and
relative permeability in two ways: by changing the pore-size distribution index, and by changing the irreducible water saturation.
Sensitivity to Variations of Pore-Size Distribution Index. Three
values of pore-size distribution index, , were considered: 0.5
(very wide range), 2 (wide range), and 4 (medium range). The
expressions used in this exercise are the drainage equations associated with the modified Brooks-Corey model, namely:
S*w =

Sw Swr
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (9)
1 Swr Sor

S*
o=

1 Sw Sor
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (10)
1 Swr Sor

Fig. 12Two-layer and isotropic formations considered for

case examples A and B, respectively. Capillary pressures, relative permeabilities, and rates of mud-filtrate invasion are assumed the same for the two layers. The two observation probes
are located within the top layer, whereas the packer flow area is
positioned within the bottom layer. Numbers located to the left
of the tool schematic indicate the distance in feet from each
probe and packer to the top of the reservoir. Reservoir dimensions and remaining formation properties are the same as those
of the base-case model.

done before, modified Brooks-Corey drainage equations were used

to define the capillary pressure and relative permeability curves.
Unlike the effect of variations of , variations of Swi dramatically
influence the simulated noise-free pressure measurements compared to those of the base-case model.
Sensitivity to Variations of Production Flow Rate
For the base-case model, the formation tester withdraws formation
fluids at a rate of 21 B/D from the packer-straddle section of the
borehole. Variations of 4 B/D to the latter production rate are
enforced by the simulator to obtain the inversion results summarized in Table 5. Note that the estimated permeability increases or

Pc,dr = PceS*
w . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (11)
2+3

krw,dr = S*
w

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (12)
2+3

2
kro,dr = S*
o 1 S*
w

, . . . . . . . . . . . . . . . . . . . . . . . . . . . (13)

where S is fluid saturation, Pc,dr is drainage capillary pressure, kr,dr

is drainage relative permeability, and Pce is capillary entry pressure. In these equations, subscripts are used to identify water (w),
oil (o), and irreducible saturations (r). The superscript * indicates
normalized saturations as defined by Eqs. 9 and 10.
Table 5 shows the corresponding inversion results. By considering that the base case value is 2, it can be concluded that this
variable has marginal impact on the final inverted properties.
Sensitivity to Variations of Irreducible Water Saturation.
Variations of the assumed irreducible water saturation were tested
for values of 0.15, 0.25, and 0.35 (used in the base-case model). As
September 2007 SPE Journal

Fig. 13Geometrical description of a finely laminated formation

model. The packer flow area is located within the bottom layer,
while the two observation probes sample pressure from the
upper layers. These layers are separated by low-permeability
intermediate layers whose boundaries and permeabilities are
known a priori. Numbers located to the left of the tool schematic
indicate the distance in feet from each probe and packer to the
top of the reservoir.
349

decreases in proportion to the decrease or increase of the rate

variation, respectively.
Sensitivity to Presence of Impermeable
Bed Boundaries
Pressure-transient measurements remain sensitive to the presence
of impermeable bed boundaries in the vicinity of measurement
points. The purpose of this section is to assess the influence of
unaccounted impermeable bed boundaries on the inversion results
for different values of layer thickness. Fig. 4 shows the finitedifference grid used to perform the simulations. The packer is
located in the middle of the formation as the layer thickness, h,
is decreased to prescribed values. There are impermeable beds

adjacent to the formation with porosity and permeability equal to

0.0001% and 0.000001 mD, respectively. Fig. 14 describes the
simulated pressure transients, and Table 6 describes the corresponding inversion results. For these cases, the inversion is performed using both drawdown and buildup pressure measurements
acquired only with one monitoring probe (probe 2) in addition to
pressure measurements acquired with the packer.
In general, the pressure differential during drawdown measured
by both the observation probe and the packer increases in the
presence of impermeable bed boundaries, thereby biasing the inversion results toward permeability values lower than that of the
base case. The closer the packer is located to an unaccounted
impermeable bed, the lower the corresponding estimate of permeability yielded by the inversion.
Sensitivity to Mud-Filtrate Invasion Parameters
This section evaluates the impact of some important assumptions
made about parameters associated with the process of mud-filtrate
invasion on the permeabilities yielded by the inversion. Specifically, three invasion parameters are given consideration:
1. Presence of an invasion zone.
2. Time of invasion.
3. Overbalance pressure.
In all these cases, the formation test is initialized using the spatial
distributions of pressure and fluid saturation derived from a previous mud-filtrate invasion simulation (i.e., we are assuming that
no invasion occurs while pressure transients are acquired by the
formation tester). This assumption may not be accurate if the formation test is performed concomitant to drilling.
Sensitivity to Presence of an Invasion Zone. Fig. 15 illustrates
the supercharging effect on pressure transients simulated for a
1-mD rock formation assuming mud and formation properties as
specified in Table 1. The original formation pressure is masked by
the large sandface pressure observed on the wellbore side of the
mud-filtrate invaded zone. Fig. 16 compares the pressure-transient

Fig. 14Effect of impermeable bed boundaries on the simulated pressure measurements for a formation of permeability
equal to 100 mD. In the figures, h designates the thickness of
the permeable rock shouldered by impermeable beds (top and
bottom bed boundaries).
350

Fig. 15Pressure supercharging effect extending more than 2

in. into the formation simulated 1.5 days after the onset of mudfiltrate invasion in an isotropic formation with permeability
equal to 1 mD. At the sample depth (grid number 29 in the
z-axis), the pressure difference between virgin formation and
sandface pressure is as high as 10 psi.
September 2007 SPE Journal

Fig. 16Pressure-transient measurements simulated to assess

the effect of presence of a mud-filtrate invaded zone for the
case of a formation with isotropic and homogeneous permeability equal to 100 mD. The two plots shown above describe
noise-free pressure measurements simulated at one of the observation probes and the packer.

measurements simulated at the two probes and the packer when the
invaded zone is included in the formation model. Relatively large
variations are observed in the simulated pressure transient measurements when the process of invasion is not considered by the
simulations.

Fig. 17Pressure measurements simulated to assess the effect

of inaccurate single-phase flow assumptions on the inverted
values of permeability for the case of a formation with isotropic
and homogeneous permeability equal to 10 mD. The two plots
shown above describe noise-free pressure measurements
simulated at one observation probe and the packer. Remaining
formation properties are the same as those of the base-case
model.

Sensitivity to Mud Overbalance Pressure. Three different values

of mud pressure were entered to the flow simulator to generate
noise-free synthetic pressure transient measurements: 3,300,
3,400, and 3,500 psi, corresponding to overbalance pressures of
300, 400, and 500 psi, respectively. Incorrect assumptions about
mud overbalance pressure (differences of approximately 100 psi)
lead to errors of 2% on the estimated permeability.

regime is inaccurately assumed to be single phase. Fig. 17 shows

the pressure-transient measurements simulated at the two probes
and the packer for the case of a formation with homogeneous and
isotropic permeability equal to 10 mD. Single-phase flow was
simulated by equating to zero the capillary pressure and water
relative permeability curves, and by assigning a value of 1 to the
oil relative permeability curve. We note a relatively large pressure
differential simulated for both types of synthetic measurements
assuming the same production flow rate. Unlike single-phase flow,
two-phase fluid flow constrains the displacement of oil by the
specific value of water saturation, thereby entailing a larger pressure differential than for the case of single-phase flow. Fig. 8,
described earlier, shows a similar comparison using a log-log plot,
where formation permeability is equal to 100 mD. Table 6 summarizes the inversion results obtained from this sensitivity analysis. It is found that inaccurate assumptions about single-phase flow
entail inverted permeabilities lower than the actual values.

Inversion Results for the Case of Single-Phase

Fluid Flow
We compare inversion results obtained with the approach developed in this paper against inversion results obtained when the flow

Discussion
The inversion algorithm includes different components that add to
the complexity of the estimation problem but, at the same time,
contribute to improving the reliability and physical consistency of

Sensitivity to Time of Invasion. Simulations of pressure transients were performed for three different times of invasion: 0.5,
1.0, and 1.5 days. The effect of unaccounted time of invasion is to
increase the inverted value of permeability when the actual time of
invasion is shorter than assumed by the inversion.

September 2007 SPE Journal

351

Fig. 18Pie chart describing the relative impact of several formation properties and formation-test parameters on the reliability of inverted values of permeability and permeability
anisotropy. The size of each slice is proportional to the standard deviation of the inverted properties compared to the corresponding target (true) property values.

the results. To emphasize such an important property of the inversion method developed in this paper, Fig. 18 summarizes the formation properties and test parameters that exhibit the largest impact on the simulated formation tester measurements. Incorrect
assumptions about viscosity are by far the most important in the
analysis (which, incidentally, is also the case for single-phase inversions). Presence of a mud-filtrate invaded zone, relativepermeability and capillary-pressure curves, knowledge of impermeable bed boundaries, and single-phase flow assumptions are
the second-largest causes of data misfit in the analysis. Moreover,
Fig. 19 suggests that the estimation error caused by additional
uncertainties associated with two-phase flow analysis is smaller
than either the error associated with the use of conventional singlephase flow techniques or the error associated with neglecting presence of mud-filtrate invasion. In other words, the two-phase character of the flow phenomenon under consideration remains crucial
to accurately interpret pressure measurements acquired with a formation tester.
Likewise, presence of permeability anisotropy causes a relatively large pressure drawdown at the packer while it decreases the
amplitude of pressure transients measured at the probes. This is not
a desirable situation for the case of high-permeability rock formations, where the amplitude of pressure transients significantly decreases and hence leads to unreliable inversions in the presence of
noise. This fact, coupled with nonuniqueness when inverting more
than one unknown parameter, requires the use of alternative strategies to constrain the solution and to redefine the model and input
measurements. Similar situations arise for the cases of unknown
petrophysical properties associated with multilayer formations. Inversion exercises emphasize the importance of good initial guesses
(obtained from auxiliary measurements such as rock-core and
well-log data) as well as physical bound constraints imposed on
the unknown properties.
Variations of fluid viscosity reveal fluid-flow characteristics
completely different from those of single-phase flow. For the basecase model, the deleterious impact of incorrect assumptions made
on oil viscosity was significant compared to that of incorrect values of water viscosity. This can be explained by the fact that the
saturation region for relative permeability and capillary pressure
fluctuates between values of water saturation of 0.45 and 0.63,
where most of the displaced fluid is oil. As inferred from the
corresponding inversion results, the effect of inaccurate assumptions about fluid viscosity is similar to that caused by inaccurate
assumptions about the character of the flow regime.
A similar behavior was observed for the case of inaccurate
assumptions about water/oil relative-permeability and capillarypressure curves. It was found that, in general, the pore-size distri352

Fig. 19Relative error in the estimation of three formation permeabilities (10, 100, and 1,000 mD). Either the assumption of
single-phase flow or the omission of mud-filtrate invasion in the
analysis leads to a much higher error than that introduced by
erroneous two-phase flow assumptions. A 0.2 perturbation of
irreducible water saturation (Swi) for the Brooks-Corey models
is probably too large, although still important in the two-phase
flow analysis for the base-case model.

bution index () associated with a modified Brooks-Corey model

marginally affected the inversion results. Significant biases in the
inversion results were observed only for the cases of highpermeability rock formations. Conversely, inaccurate assumptions
on irreducible water saturation (Swi) have a measurable impact on
the estimated values of permeability because of the increase of
effective permeability to water in relation to effective permeability
to oil.
The sensitivity analysis to presence of impermeable beds was
intended to assess the effects of inaccurate assumptions about the
distance to upper and lower impermeable beds on the inverted
values of permeability. Presence of unaccounted impermeable
beds near the sampling points caused a relatively large increase of
pressure drawdown measurements at the packer and monitoring
probes. Consequently, the inversion algorithm yielded values of
permeability lower than the target values.
In field applications, presence of multiple local minima could
be overcome in a similar manner as described in this paper with the
use of restart sequences. In addition, initial values of unknown
parameters could be obtained from independent sources of information (e.g., well logs, single-phase transient analysis, and rockcore data) and used to enforce bound constraints on the inversion.
Conclusions
The following is a summary of the most important conclusions
stemming from this paper:
1. We showed that significant variations of pressure can be caused
by inaccurate assumptions made about two-phase rock-fluid
properties (such as relative permeability, capillary pressure, and
oil viscosity), depending on specific conditions of invasion and
values of permeability. Traditional approaches used for the interpretation of dual-packer pressure measurements are based on
the assumption of single-phase flow. To estimate permeability,
these approaches include a correction factor to account for
two-phase flow effects. The algorithm developed in this paper
explicitly considers the two-phase nature of fluid flow during
the acquisition of formation-tester measurements. Moreover, we
have shown that neglecting the physics of two-phase flow and
mud-filtrate invasion can result in permeability estimation errors
as high as 100%.
2. By explicitly including the processes of dynamic mudcake
growth and mud-filtrate invasion into the inversion algorithm,
the methodology presented here offers the advantage of not
September 2007 SPE Journal

being restricted to assume stationary and/or piston-like mudfiltrate invaded layers. Once initial pressures and fluid saturations (calculated from the simulation of mud-filtrate invasion)
are explicitly included as initial conditions for the simulation of
formation-tester measurements, there is no need to include arbitrary fluid interfaces. Fluid saturation is allowed to change in
the vicinity of the packer during the test. This flexibility in
the simulation improves the estimation of permeability, especially in cases wherein significant fluid cleanup is observed
during the test.
3. The combination of numerical near-wellbore simulations with a
computationally efficient inversion algorithm based on the Levenberg-Marquardt minimization method provides a systematic
way to reduce nonuniqueness in the presence of noisy pressure
measurements. We showed that a priori information on the
unknown parameters is necessary to reduce nonuniqueness.
4. Synthetic pressure measurements were considered to appraise
the reliability of the new inversion method proposed in this
paper for the interpretation of wireline formation-tester measurements acquired with dual-packer modules. Although this
approach still needs to be tested with field data, it provided
reliable inversion results for the various synthetic formation
models considered in this paper.
Nomenclature
e(x)
h
IM
k
kh
kh /kv
kro
k0ro
krw
k0rw
kv
M
nr
nz
N
p
pmj
psj
p
Pc
Pce
Pd
q
re
rw
So
Sor
S*
o
Sw
Swi, Swr
S*
w
x
Wd

o
w

cost function
vector of residuals
formation thickness, ft
impact value
absolute permeability, mD
horizontal permeability, mD
permeability anisotropy ratio
oil-phase relative permeability
kro endpoint, mD
water-phase relative permeability
krw endpoint, mD
vertical permeability, mD
number of measurements
grid number in the radial direction
grid number in the vertical direction
number of unknowns
pressure prior to formation test, psi
measured pressure data, psi
simulated pressure data, psi
pressure differential, psi
capillary pressure, psi
capillary entry pressure, psi
particle diameter for the formation rock, m
fluid flow rate, B/D
external radius, ft
wellbore radius, ft
oil-phase saturation
residual oil saturation
normalized oil-phase saturation
water-phase saturation
irreducible water-phase saturation
normalized water-phase saturation
estimate of the model parameter
data weight matrix
pore size distribution index
Lagrange multiplier
oil-phase viscosity, cp
water-phase viscosity, cp
standard deviation
estimators covariance matrix
effective porosity, fraction

September 2007 SPE Journal

C(x)

Acknowledgments
The authors thank Kamy Sepehrnoori for his assistance during the
development of the two-phase fluid-flow algorithm. Our gratitude
is also extended to three anonymous reviewers whose constructive
technical and editorial feedback improved the quality of the first
manuscript. The work reported in this paper was funded by the
University of Texas at Austins Research Consortium on Formation Evaluation, jointly sponsored by Anadarko, Aramco, Baker
Atlas, BP, British Gas, ConocoPhillips, Chevron, ENI E&P, ExxonMobil, Halliburton, Hydro, Marathon, the Mexican Institute for
Petroleum, Occidental Petroleum Corporation, Petrobras, Schlumberger, Shell International E&P, Statoil, Total, and Weatherford.
References
Alpak, F.O. 2005. Algorithms for Numerical Modeling and Inversion of
Multi-phase Fluid Flow and Electromagnetic Measurements. PhD dissertation. Austin, Texas: University of Texas at Austin.
Angeles, R. 2005. Inversion of Permeability and Permeability Anisotropy
from Straddle-Packer Formation Tester Measurements using the Physics of Two-Phase Immiscible Flow and Invasion. MSc thesis. Austin,
Texas: University of Texas at Austin.
Ayan, C. et al. 2001. Characterizing Permeability With Formation Testers.
Oilfield Review 13: 223.
Badaam, H., Al-Matroushi, S., Young, N., Ayan, C., Mihcakan, M., and
Kuchuk, F.J. 1998. Estimation of Formation Properties Using Multiprobe Formation Tester in Layered Reservoirs. Paper SPE 49141 prepared for presentation at the SPE Annual Technical Conference and
Exhibition, New Orleans, 2730 September. DOI: 10.2118/49141-MS.
Brooks, R.H. and Corey, A.T. 1964. Hydraulic properties of porous media.
Colorado State University Hydrology Paper 3. Fort Collins, Colorado:
Colorado State University Press.
Coates, G.R. and Denoo, S. 1981. The Producibility Answer Product. The
Technical Review 29 (2): 5463.
Cramr, H. 1945. Mathematical Methods of Statistics. Princeton, New
Jersey: Princeton University Press.
Gk, I.M., Onur, M., Hegeman, P.S., and Kuchuk, F.J. 2006. Effect of an
Invaded Zone on Pressure-Transient Data From Multiprobe and
Packer-Probe Wireline Formation Testers. SPEREE 9 (1): 3949. SPE84093-PA. DOI: 10.2118/84093-PA.
Habashy, T.M. and Abubakar, A. 2004. A General Framework for Constraint Minimization for the Inversion of Electromagnetic Measurements. Progress in Electromagnetics Research 46: 265312.
Hawkins, M.F., Jr. 1956. A Note on the Skin Effect. Trans., AIME, 207:
356357.
Hurst, S.M., McCoy, T.F., and Hows, M.P. 2000. Using the Cased-Hole
Formation Tester Tool for Pressure Transient Analysis. Paper SPE
63078 presented at the SPE Annual Technical Conference and Exhibition, Dallas, 14 October. DOI: 10.2118/63078-MS.
Jackson, R.R., Banerjee, R., and Thambynayagam, R.K.M. 2003. An Integrated Approach to Interval Pressure Transient Test Analysis Using
Analytical and Numerical Methods. Paper SPE 81515 presented at the
SPE Middle East Oil Show, Bahrain, 912 June. DOI: 10.2118/81515MS.
Kuchuk, F.J. 1998. Interval Pressure Transient Testing With MDT PackerProbe Module in Horizontal Wells. SPEREE 1 (6): 509518. SPE53002-PA. DOI: 10.2118/53002-PA.
Lee, H., Wu, Z., and Torres-Verdn, C. 2004. Development of a TwoDimensional, Axi-Symmetric Single Well Code for Two-Phase Immiscible Fluid Flow, Salt Mixing, and Temperature Equilibration in the
Near Wellbore Region With Applications to the Simulation of MudFiltrate Invasion And Formation Tester Data. Appendix II. Fourth Annual Report. Austin, Texas: Formation Evaluation Program at University of Texas at Austin.
Leverett, M.C. 1941. Capillary Behavior in Porous Solids. Trans., AIME,
142: 152169.
Onur, M., Hegeman, P.S., and Kuchuk, F.J. 2004. Pressure-Transient
Analysis of Dual Packer-Probe Wireline Formation Testers in Slanted
Wells. Paper SPE 90250 presented at the SPE Annual Technical Conference and Exhibition, Houston, 2629 September. DOI: 10.2118/
90250-MS.
353

Pop, J.J., Badry, R.A., Morris, C.W., Wilkinson, D.J., Tottrup, P., and
Jonas, J.K. 1993. Vertical Interference Testing With a WirelineConveyed Straddle-Packer Tool. Paper SPE 26481 presented at the
SPE Annual Technical Conference and Exhibition, Houston, 36 October. DOI: 10.2118/26481-MS.
Proett, M.A., Chin, W.C., and Mandal, B. 2000. Advanced Dual Probe
Formation Tester With Transient, Harmonic, and Pulsed Time-Delay
Testing Methods Determines Permeability, Skin, and Anisotropy. Paper SPE 64650 presented at the SPE International Oil and Gas Conference and Exhibition in China, Beijing, 710 November. DOI:
10.2118/64650-MS.
Pucknell, J.K. and Clifford, P.J. 1991. Calculation of Total Skin Factors.
Paper SPE 23100 presented at the SPE Offshore Europe Conference,
Aberdeen, 36 September. DOI: 10.2118/23100-MS.
Wu, J., Torres-Verdn, C., Proett, M.A., Sepehrnoori, K., and Belanger, D.
2002. Inversion of Multiphase Petrophysical Properties Using Pumpout
Sampling Data Acquired With a Wireline Formation Tester. Paper SPE
77345 presented at the SPE Annual Technical Conference and Exhibition, San Antonio, Texas, 29 September2 October 2. DOI: 10.2118/
77345-MS.
Xian, C., Carnegie, A., Al Raisi, M.R., Petricola, M., and Chen, J. 2004.
An Integrated Efficient Approach To Perform IPTT Interpretation. Paper SPE 88561 presented at the SPE Asia Pacific Oil and Gas Conference and Exhibition, Perth, Australia, 1820 October. DOI: 10.2118/
88561-MS.
Zeybek, M., Ramakrishnan, T.S., Al-Otaibi, S.S., Salamy, S.P., and Kuchuk, F.J. 2001. Estimating Multiphase Flow Properties Using Pressure
and Flowline Water-Cut Data From Dual Packer Formation Tester
Interval Tests and Openhole Array Resistivity Measurements. Paper
SPE 71568 presented at the SPE Annual Technical Conference and
Exhibition, New Orleans, 30 September3 October. DOI: 10.2118/
71568-MS.

AppendixPetrophysical Correlations for

Physically Consistent Inversion
Properties such as permeability are changed as the nonlinear inversion method approaches the minimum of the quadratic cost
function. This variation of rock-formation properties cannot be
performed without the use of physically consistent correlations
between relevant petrophysical properties. The purpose of this
Appendix is to describe the petrophysical correlations enforced by
the inversion algorithm to honor transient pressure measurements.
Relative permeabilities are assumed constant for a given type
of rock regardless of permeability. Capillary pressures and porosity, however, usually exhibit a strong correlation with permeability
and therefore need to be automatically adjusted by the inversion.
For the case of porosity, the algorithm first uses average values
of irreducible water saturation (Swi) and porosity (b) at the pressure sensor depth (usually known a priori from well logs) to obtain
a base permeability (kb). The correlation model used to enforce
a quantitative relationship between irreducible water saturation,
porosity, and permeability is the one proposed by Coates and
Denoo (1981), namely,

kb = 100

2b1 Swi 2
, . . . . . . . . . . . . . . . . . . . . . . . . . . . . (A-1)
Swi

where kb is given in mD.

Subsequently, the algorithm makes use of the Blake-Kozeny
model to determine a base particle diameter (pdb). In metric units,

354

pdb =

1501 b2kb

3b

. . . . . . . . . . . . . . . . . . . . . . . . . . . . (A-2)

Using this expression, porosity () can be consistently scaled for

a given value of permeability (k) using the recursive formula

= 150

k1 2
p2db

13

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . (A-3)

For capillary pressure, a similar procedure is employed by combining the Leverett J-function (Leverett 1941) and a modified
Brooks-Corey (1964) reference model. Here, the pore-size distribution coefficient () is assumed constant for a given rock.
Renzo Angeles is a graduate research assistant and a PhD
candidate at the University of Texas at Austin. From 2000 to
2003, he worked as a field engineer for Schlumberger, receiving training in Peru, Colombia, Ecuador, the United States, and
Canada. He is a recipient of the 2003 Presidential Endowed
Osmar Abib Scholarship and the 200506 Chevron Scholarship.
His interests involve pressure transient analysis, numerical simulation of reservoirs, inversion and optimization techniques, and
formation characterization. His PhD research focuses on the
quantitative analysis and inversion of formation tester measurements acquired in highly deviated wells using two- and
three-phase flow analysis including the effects of mud-filtrate
invasion. Carlos Torres-Verdn has been with the Department
of Petroleum and Geosystems Engineering of University of
Texas at Austin, where he currently holds the position of associate professor, since 1999. From 1997 to 1999, he was Reservoir
Specialist and Technology Champion with YPF (Buenos Aires,
Argentina). Since 1999, he has conducted research on borehole geophysics, well logging, formation evaluation, and integrated reservoir characterization. He is corecipient of the 2003,
2004, and 2005 Best Paper Award by Petrophysics, and he is
the recipient of SPWLAs 2006 Distinguished Technical Achievement Award. He holds a PhD degree in engineering geoscience from the University of California, Berkeley. Hee-Jae Lee
is currently a PhD candidate in the Petroleum and Geosystems
Engineering Department of the University of Texas at Austin. He
is working in the Formation Evaluation Research Group supervised by Torres-Verdn. His main research area is in geomechanical fluid-flow coupling, near-wellbore simulation, and is
the main code developer of the University of Texass Formation
Evaluation Tool Box (UTFET). He holds BSc and MSc degrees in
petroleum engineering from Hanyang University, Korea. Faruk
O. Alpak is a research reservoir engineer with Shell International E&P at Bellaire Technology Center, Houston. His research
interests include parallel reservoir-simulation techniques, computational fluid dynamics, uncertainty analysis, inverse problems, numerical optimization, and electromagnetic wave
propagation. Alpak holds PhD and MSc degrees in petroleum
engineering from University of Texas at Austin. James Sheng is
currently working for Total E&P in Houston as Research Adviser.
Previously, he was Lead Scientist with Baker Hughes. He also
worked as a reservoir engineer with several major and national
oil companies. His work experience includes reservoir simulation and numerical modeling, well testing, wireline testing and
sampling, heavy-oil recovery, production forecast, and enhanced oil recovery. He holds a PhD degree from the University of Alberta. He received several professional awards including the Outstanding Technical Editor Award for SPEREE (2005)
and the Best Paper Award in JCPT (1998). He is currently serving
as a Review Chairperson for SPEREE.

September 2007 SPE Journal